Price Discrimination∗

Mark ArmstrongDepartment of EconomicsUniversity College London

October 2006

1 Introduction

In broad terms, one can say that price discrimination exists when two “similar” productswhich have the same marginal cost to produce are sold by a firm at different prices.1 Thispractice is often highly controversial in terms of its impact on both consumers and rivals.This chapter aims to explain some of the main economic motives for price discrimination,and to outline when this practice will have an adverse or beneficial effect on consumers, rivalsand on total welfare.Broadly speaking, there are three main reasons why competition policy may be concerned

with price discrimination. First, a dominant firm may “exploit” final consumers by means ofprice discrimination, with the result that total and/or consumer welfare are reduced. Here,the question that needs to be addressed is: in what circumstances does price discriminationby a dominant firm have an adverse effect on welfare? The answer to this, as will be clear inthis survey, often depends on the welfare standard that guides the application of competitionlaw. However, in practice competition authorities hardly ever concern themselves with pricediscrimination as an exploitative device.Second, and especially in Europe, it is sometimes a policy objective to attain a “single

market” across the region. Arguably, one manifestation of a single market is that a firmdoes not set different prices in different regions, or at least it does not prevent arbitrageurs

∗I am very grateful to Paolo Buccirossi and John Vickers for comments. More technical and detaileddiscussions of some of the material presented here is presented in Armstrong (2006). In preparing thissurvey I have benefited from consulting the two earlier surveys by Hal Varian (1989) and Lars Stole (2006).

1There seems to be no consensus on a precise definition. Stigler (1987) suggests a definition that applies toa wider class of cases: discrimination exists when two similar products are sold at prices that are in differentratios to their marginal costs. Which of these definitions we use makes no difference for the purposes of thispaper.

1

reselling goods sourced in the low-price region to the high-price region. That is to say, firmsare often prevented from segmenting markets with a view to engaging in price discrimination.In Europe, this concern has lead to a very hostile attitude by the authorities to attempts byfirms to prevent “parallel imports”, for instance.Third, and perhaps most importantly from a competition authority’s point of view, we

may be concerned that price discrimination can be used by a dominant firm to “exclude”(or weaken) actual or potential rivals. The question is in which cases price discriminationcan be an effective way to put the buyer’s rivals (primary line injury) or the seller’s rivals(secondary line injury) at disadvantage so as to force them to exit the market, or inducethem to compete less aggressively.The appropriate antitrust treatment of price discrimination may require consideration of

more than one of these concerns. For instance, a form of price discrimination may poten-tially be an efficient way to supply services to final consumers, and yet it may also possessexclusionary effects. In such cases, competition law and policy needs to balance the risk ofpreventing firms from pricing their products efficiently with the risk of permitting conductthat leads to a less competitive market structure.This chapter aims to explain some of the main economic motives for price discrimination,

and to outline when this practice will have an adverse or beneficial effect on consumers,rivals and total welfare. The plan of the chapter is as follows. Section 2 outlines some ofthe principal methods of price discrimination. Section 3 shows how the ability to engagein price discrimination can sometimes lead to efficient (e.g., marginal cost) prices, whichclearly leads to welfare gains. Section 4 discusses how price discrimination can open up newmarkets or shut down existing markets. Section 5 examines when price discrimination causestotal output to rise or fall. Section 6 discusses when the introduction of price discriminationwill cause some prices to rise and other to fall, while sections 7 and 8 focus on less familiarsituations in which price discrimination causes all prices to fall or all prices to rise. Section9 examines the impact of price discrimination on entry incentives. Section 10 introducesdynamic price discrimination, while section 11 outlines the impact of price discrimination invertically-related markets.The discussion throughout focusses on the underlying economics of price discrimination

and its impact on profits, entry, consumer surplus, and welfare. Where relevant, though,particularly prominent anti-trust cases are mentioned as we go along.2

2 Forms of price discrimination

There are numerous business practices which fall under the heading of price discrimination.First, consider static situations in which consumers buy all relevant products in a single

2Section 3.7 of Varian (1989) summarizes the origins of the legal approaches to price discrimination in theUnited States, which initially focussed on protecting small retailers against large chain stores. For a surveyon the application to price discrimination of Article 82 of the EC Treaty, see Geradin and Petit (2005).

2

period. In most markets, firms set the charge for purchase of their products by means ofa simple price per unit for each product, where these prices do not depend on who makesthe purchase. Such tariffs (i) are anonymous (they do not depend on the identity of theconsumer), (ii) do not involve quantity discounts for a specific product (i.e., there are no“intra-product” discounts), and (iii) do not involve discounts for buying a range of products(i.e., there are no “inter-product” discounts). Various forms of price discrimination are foundby relaxing restrictions (i)—(iii).

Non-anonymous price discrimination: This occurs when a firm offers a different tariff toidentifiably different consumers or consumer groups. When the tariff also involves simpleper-unit pricing (rather than nonlinear pricing or bundling), this is the familiar case of third-degree price discrimination. Examples of this practice include selling the same train ticketat a discount to senior citizens, selling the same car at different prices in two countries,or selling the same drug at difference prices for human and animal use. Unless arbitragebetween consumer groups is very easy or competition between firms is almost perfect, weexpect that any firm, if permitted to do so, would wish to set different tariffs to differentgroups. In antitrust cases this type of price discrimination can occur when the alleged abuseconsists in “selective price cuts” or “geographic price discrimination”.3

Another example of this form of price discrimination is first-degree price discrimination,where each consumer is charged exactly her willingness-to-pay for the product(s). In itspurest form, the information needed for first-degree price discrimination makes it more of atheoretical benchmark than a realistic business strategy. However, it provides a transparentlimit framework in which to discuss the possible efficiency gains from price discrimination,as well as its distributional impact on consumers.

Quantity discounts: This occurs when the per-unit price for a specific product decreasesas the number of purchased units increases. A simple–and easy implemented–instance ofthis is a two-part tariff, whereby a buyer must pay a fixed charge in return for the rightto purchase any quantity at a constant marginal price. There are two distinct motives touse nonlinear tariffs. First, nonlinear tariffs provide a more efficient means by which togenerate consumer surplus. With linear pricing, the only way to make profit is to set pricesabove costs, which entails deadweight losses. With a two-part tariff, however, a firm canextract profit using the fixed charge, while leaving marginal prices close to marginal costs(which then maximizes the size of the “pie” to be shared between consumer and firm). Thisrole for nonlinear pricing exists even if all consumers are similar. A second role emerges ifconsumers have heterogeneous tastes for a firm’s products. In this case, a nonlinear tariffcan be used to sort different types of consumers endogenously (so-called second-degree price

3Prominent antitrust cases in the European Union concerning selective price cuts and geographic pricediscrimination are Irish Sugar and Compagnie Maritime Belge, for the first category, and United Brandsand Tetra Pak II for the second.

3

discrimination).4 If some consumers gain higher utility from the product than others, then ifthe firm offers a tariff where the marginal price declines with volume, it will typically makehigher profit than if it offers the same marginal price to all consumers.5

Pure quantity discounts are generally not challenged by competition authorities if theymerely reflect cost efficiencies stemming from the larger volume of product sold (and aretherefore not discriminatory). However, at least in Europe, there is hostility towards dis-counts offered by a dominant firm which do not reflect costs.6 This approach, which seemsto be overly rigid, does not recognize the efficiencies that may stem from the pricing method,independent of the cost efficiencies related to the scale of the transaction.

Bundling discounts: This occurs when the price for one product is reduced if the consumeralso buys another product. Two variants of bundling exist: (i) pure bundling is where aconsumer can only purchase the products as bundle and there is no scope for buying anindividual item, and (ii) mixed bundling, where the firm sets prices for a bundle and also forindividual items. In general, unless products have negligible marginal costs or are perfectcomplements, pure bundling is a rather inefficient business practice. The reason is that itforces some consumers to purchase products for which their willingness-to-pay is smaller thanthe cost of supply. Thus, we expect that “all inclusive holidays” will often be an inefficientway to market holidays, since efficiency requires that consumers only consume what theyvalue. In addition, we will see that pure bundling can provide a means by which to deter entryby single-product firms. Mixed bundling (with two products) sorts consumers endogenouslyinto three groups: those with a strong taste for both products (who buy the bundle from thefirm), those with a strong preference for product 1 but weak preferences for product 2 (whobuy just product 1), and those with the reverse tastes (who buy just product 2). Mixedbundling is closely related to two-part pricing: the “first” item from the firm is expensive,while the “second” is relatively cheap. Indeed, since it is often hard to pin down what itmeans for products to be “distinct”, in practice it may be hard to distinguish (intra-product)quantity discounts from (inter-product) bundling discounts. (Should a season ticket for aconcert series, say, count as a quantity discount for purchasing several units of the sameproduct, or as a discount for purchasing several distinct items?)

Next turn to dynamic situations. There are several ways in which a firm can set different

4However, there are plenty of examples of second-degree discrimination that are not to do with nonlinearpricing. For instance, many retailers use coupons placed in newspapers to segment the market. It is plausiblethat those consumers who take the trouble to cut out and use a coupon will also have more elastic demand,and should therefore face a lower price. See Narasimhan (1984), for instance.

5See Tirole (1988, section 3.5.1), for instance.6In its judgment in Michelin II the European Court of First Instance suggested that if discounts are

not based on cost efficiencies they should be regarded as in beach of Article 82. For further discussion ofcompetition law and policy towards quantity discounts and related practices, see Vickers (2005, section 2.4)and the “Syposium on Loyalty Rebates”, Competition Policy International, volume 1, number 2, Autumn2005.

4

prices for essentially the same product over time. If consumers wish to buy a single item atsome point in time (e.g., a new novel), then a firm might be able to make those consumerswith a higher reservation price pay more than other consumers by setting a price for the itemwhich decreases over time. This inter-temporal price discrimination is essentially anotherform of second-degree discrimination. A crucial difference between static and dynamic formsof price discrimination is that, in the latter case, a firm may not have the ability to committo future prices. That is to say, once a firm has sold its product to the initial (enthusiastic)pool of consumers, it typically has an incentive to reduce its price to extract profit from theremaining, lower-value consumers. Indeed, in simple models, it turns out that the firm wouldprefer not to have the ability to price discriminate in this way, but it cannot help but offerdeclining prices over time. This is the essence of the famous Coase (1972) problem facedby firms selling to forward-looking buyers. In these situations, policy which forbids pricediscrimination may have the effect of endowing the firm with commitment power, with theresult that all prices can rise. Thus, it is perhaps not surprising that competition authoritieshave rarely investigated this type of dynamic price discrimination.Other issues arise if consumers wish to purchase a product repeatedly (e.g., from a su-

permarket or an online retailer). In this case, a firm may be able to base its price todayon whether (or how much, or what) a consumer has purchased from it in the past. Thisbehaviour-based price discrimination is becoming increasingly prevalent due to the improvedtechnological ability to track consumer behaviour by means of the internet, loyalty cards,and so on. This kind of price discrimination is a dynamic variant of nonlinear pricing, andone which raises interesting and subtle issues. However, one difference with static nonlinearpricing is that, in many cases, a firm does not announce its discounting strategy. Partlythis is because it would often be too complicated to describe its strategy–for instance, asupermarket’s personalized vouchers to a particular consumer will depend upon his historyof purchases, and this will be impractical to communicate in detail. But partly there isthe commitment problem again. For instance, a supermarket may notice that a past cus-tomer appears to have starting shopping elsewhere (his loyalty card has not been used, forinstance), and it may then decide to mail the consumer a financial inducement to return tothe shop. It is unlikely that the supermarket will wish to publicize such a strategy. In sum,information about shopping habits presents firms with an ability to set “personalized” (i.e.,discriminatory) prices. It is intrinsically hard to commit to such personalized prices.7

The final class of discriminatory pricing discussed involve the pricing of inputs to down-stream firms. An issue of frequent anti-trust concern is the wholesale price a vertically-integrated firm should be permitted to charge a downstream rival. When the firm sets toohigh a wholesale price (relative to its own retail price), the firm might be said to “discrim-

7However, the information and commitment problems are sometimes less severe when the buyers aredownstream firms with which the seller has long-term relationships. In these cases the dominant seller mayoffer what in the antitrust jargon are sometimes referred to as “target rebates”. This practice conditionsthe rebate to the meeting of a threshold based on the past purchases of the buyer. Target rebates have beeninvestigated in the cases Michelin I and II, British Airways and Irish Sugar.

5

inate” against its rival. The interpretation of “too high” is not obvious, however, and thisissue is discussed later in the chapter. A second issue is whether a vertically-separate up-stream supplier should be able to discriminate in its wholesale contracts among downstreambuyers. A major difference between supplying an input to downstream firms and supply-ing a product to final consumers is that, in the former case, the contracts are often muchmore complex and “personalized” than those typically offered to final consumers. As such,the contract with one downstream firm might not be known to a rival downstream firm.This “secret deals” problem again raises the issue of credibility. We will see that if theupstream supplier can offer discriminatory deals to downstream firms, it may be forced tooffer generous deals which ultimately benefit final consumers. But if public policy forbidssuch discriminatory behaviour, this may restore the monopolist’s ability to set high pricesto its downstream buyers.Regardless of the method of price discrimination, it is necessary that consumer arbitrage

not unravel the discriminatory prices. For instance, if a firm wishes to set a lower price in onecountry it is important that consumers in higher-price countries not be able easily to importthe same product from the low-price country. (Prominent examples of price discriminationby country include cars and pharmaceuticals.) Similarly, a season ticket holder should not beable to let others use the season ticket, or consumers should not easily be able to pretend tobe new customers at a firm in order to take advantage of its introductory offers. Therefore,when policy makers wish to discourage price discrimination, they will often take the indirectroute of ensuring that consumer arbitrage is as easy as possible. For instance, as mentionedearlier, European competition law is very hostile to firms preventing parallel imports of theirproducts when those firms are dominant or enter into anti-competitive agreements with otherfirms.8

3 Price discrimination can lead to efficient prices

In many cases, the welfare problems caused by firms exploiting their market power are due tofirms having insufficient information about their consumers’ preferences, or being constrained(by public policy, for instance) in their ability to condition prices on their information aboutconsumers. In some circumstances, allowing firms to engage in price discrimination canimplement efficient prices. In these cases, total welfare is unambiguously improved, althoughthe impact on consumers may be negative.One familiar example of this is first-degree discrimination, where a monopolist has perfect

8An important recent case is case C-53/03 Syfait v GlaxoSmithKline before the European Court of Justice.This involved a pharmaceutical company wishing to prevent its products, which are sold at a low price inGreece, from being re-imported into high-price countries. Advocate General Jacobs issued an opinion inOctober 2004 that such constraints on parallel imports in this case should not necessarily be consideredabusive since the price differentials stemmed from state intervention in the prices for drugs in each country.

6

information about each consumer’s willingness-to-pay for its product(s).9 To be concrete,suppose there is a population of consumers, each of whom wishes to consume a single unitof the firm’s product. Willingness-to-pay for this item is denoted v and this varies amongconsumers according to the distribution function F (v). Thus, if a consumer has valuationv and faces the price p, he will buy if v ≥ p. Suppose the firm has unit cost c. If pricediscrimination is not permitted (or is otherwise not possible), the firm will choose its singleprice p to maximize profit (p−c)(1−F (p)). Clearly, the chosen price will be above cost, andtotal surplus is not maximized. (It is efficient to serve all those consumers with v ≥ c, butonly those with v ≥ p > c are served.) If the firm can somehow observe each consumer’s v andis permitted to discriminate on that basis, it will charge each consumer the maximum possible(i.e., p = v) provided this price covers its cost of supply. In other words, an efficient outcomeis achieved. However, the firm appropriates the entire gains from trade and consumers areleft with nothing. Thus, the benefits of allowing first-degree price discrimination depend onthe chosen welfare standard: with a total welfare standard such discrimination is beneficial,whereas with a consumer standard it is not. Since it is rather common that the impact ofprice discrimination on consumers is the opposite to its impact on overall welfare, this issue–that policy towards price discrimination depends on the chosen welfare standard–appearsrepeatedly in the price discrimination literature.Another example of first-degree price discrimination involves two-part tariffs. Suppose

that the monopolist knows the utility each consumer gains from its product. Specifically,suppose that a particular consumer has surplus u(q) − T if she consumes q units of theproduct in return for a payment T . Consumers must receive a non-negative surplus if theyare to buy from the firm at all. If the firm has a unit cost c its profit-maximizing strategyis to maximize T − cq subject to the consumer’s participation constraint u(q) − T ≥ 0.This entails choosing q to maximize u(q) − cq so that total surplus from the interactionis maximized. This outcome can be implemented by means of a two-part tariff: the firmsets the marginal price equal to marginal cost c and sets the fixed charge to extract all theconsumer’s surplus. (This fixed charge will differ from consumer to consumer, dependingon their preferences u(·).) Again, this results in the efficient level of consumption, whileconsumers are left with no surplus.10

Monopoly first-degree price discrimination is merely an extreme form of a fairly commonsituation. Lack of information about consumer tastes, in combination with market power,leads to welfare losses as a firm faces a trade-off between volume of demand and the profit itmakes from each consumer. In many cases, if the firm can price discriminate more finely it

9Armstrong (1999) and Bakos and Brynjolfsson (1999) show how a monopolist supplying many productscan sometimes be in a position to practise (approximate) first-degree price discrimination, even if it does notknow the precise willingness-to-pay for any individual item.10Price discrimination can also lead to efficient supply to some consumers, if not all. For instance, in

standard models of nonlinear pricing, a profit-maximizing monopolist will ensure that those consumers withthe strongest tastes for its product will face a marginal price equal to the firm’s marginal cost. See Willig(1978), for example.

7

will be able to extract consumer surplus more efficiently, and this will often lead to greateroverall welfare. However, it is consumers’ private information that protects them againstgiving up their surplus to a monopoly. Therefore, when a price-discriminating monopolisthas improved information about its consumers this will often lead to a reduction in consumersurplus.Competition between suppliers is another means by which consumers are protected

against surplus extraction. Therefore, even if firms know everything about a consumer’stastes, competition ensures the consumer will still be left with surplus. In competitiveenvironments, whether consumers are better or worse off when firms can practice price dis-crimination is a subtle question, as we will see throughout this survey.Consider for example the effects of firms offering two-part tariffs instead of linear prices.

(This discussion assumes that consumers buy all supplies from one firm or the other, i.e.,there is “one-stop shopping”.) With linear pricing, firms’ prices will be close to their marginalcosts if the market is competitive, and prices will be higher when the firms have more marketpower. When firms offer two-part tariffs their marginal prices will usually be lower thanwhen linear prices are employed. For instance, in the special case where all consumers havethe same demand function, firms will sets marginal prices exactly equal to marginal cost,since that is the most profitable method for a firm to deliver a particular level of consumersurplus. Total welfare often increases if two-part tariffs are used instead of linear prices,since the marginal price falls to cost. With more intricate analysis one can also show thatprofit increases with this form of price discrimination, while consumers are typically worseoff.11 Thus, this competitive setting resembles the monopoly setting with two-part tariffsjust discussed: welfare and profits increase but consumers are harmed by the use of two-parttariffs. The main effect of competition here is that consumer surplus is no longer driven downto zero when two-part tariffs are used. To confuse the issue, though, we will see alternativesituations in section 7 where the reverse happens: when competing firms know everythingabout consumer tastes and price accordingly, firms are harmed and all consumers are betteroff.Finally, in the context of regulated monopoly, socially optimal prices almost always exhibit

price discrimination. Ramsey prices–the prices which maximize welfare subject to theregulated firm covering its costs, including fixed costs–depend on demand conditions inmuch the same way as an unregulated monopolist’s prices do. For instance, when the firmserves a number of independent markets, each with the same marginal cost of supply, Ramseyprinciples suggest that the most efficient way to cover the firm’s production cost is to seta higher price in those markets where consumer demand is less elastic, exactly as would bethe case with an unregulated profit-maximizing firm. In sum, socially optimal prices arediscriminatory whenever the regulated firm has fixed costs of operation which need to befunded by price-cost markups.

11See Corollary 1 in Armstrong and Vickers (2001) and Yin (2004). Armstrong and Vickers (2006) showthat these results extend to many situations where consumers have heterogenous demand functions.

8

4 Price discrimination can open (or shut) markets

It is possible that permitting price discrimination will open markets that would otherwise notbe served at all. For instance, suppose a monopolist faces two independent markets, one ofwhich is “high value” and the other is “low value”. When discrimination is allowed, supposethe discriminatory price in the high-value market is significantly higher than the choke pricewhich causes demand in the low-value market to fall to zero. Then, if the size of the high-value market is sufficiently large compared to the low-value market, when discrimination isnot allowed the monopoly will choose to serve only the high-value market. In such cases,granting permission to discriminate results in a Pareto improvement: the strong market’sprice is unchanged while the weak market is served, which increases the surplus of consumersin the weak market as well as the firm’s profit.12

A simple example of this phenomenon is the following. There are two kinds of consumer:consumers in market 1 each wish to consume a single unit of the product and are willing topay up to 4 for this unit, while consumers in market 2 each wish to consume a unit and arewilling to pay 2 for this unit. Suppose the monopoly has no cost of production. In this case,if the firm must charge the same price to both markets it will choose to set the price 4 andserve only market 1 whenever there are more consumers in that market than in market 2. (Ifthere are more consumers in market 2 then the firm will prefer to set the price 2 and serveall consumers.) If price discrimination is possible then the firm will set a price 4 in market1 and a price 2 in market 2, and this results in a Pareto improvement.13

Another aspect of this issue is that the higher profit which price discrimination generatescan provide a more effective way for a firm to cover its fixed costs. In some cases, a monopolymight be profitable overall if it can price discriminate but unprofitable if it cannot. Abroadcaster, for instance, might only be profitable if it can bundle its channels together, ora rail operator might only be able to break even if it can discriminate between low incomeand other travellers. In such cases, all markets would be shut down if price discriminationwere not allowed.Further examples of how the introduction of price discrimination might open or shut a

market involve nonlinear pricing. Suppose there are two groups of consumers, high-valueusers and low-value users, and the firm cannot distinguish between the two groups directly.

12See Layson (1994) for formal analysis along these lines. We follow the literature and say that a marketwhere the price rises with discrimination is a “strong” market while a market where the price falls is a “weak”market.13Note that a variant of this example shows that price discrimination can sometimes cause markets to

shut down. Suppose there is a third intermediate market where consumers have a valuation of 3 for aunit. Suppose 50% of consumers are in market 1, 10% are in market 3 and 40% are in market 2. Withoutprice discrimination, one can show that the best strategy for the firm is to set a price of 2 and to serve allconsumers. Suppose next that the firm can distinguish only two groups of consumers, those in market 1 andthose who are in either market 2 or 3. One can show that the optimal strategy for the firm is set a price of2 to those in market 1 and to set a price of 4 to those in market 2/3. Thus, permitting price discriminationcauses the firm to abandon market 3, and this causes welfare to fall.

9

Both types of consumer potentially wish to consume up to two units of the product. Low-value users have utility of 2 from their first unit and further utility of 1 from the secondunit. High-value users have utility twice as high: their utility is 4 from the first unit and2 from the second unit. The fraction of high-value users among the consumers is λ, say. Ifproduction is costless, when the firm sets a linear price the profit-maximizing price is p = 2(regardless of the fraction λ). With such a price, the low-value users consume a single unitwhile the high-value users consume two units. Suppose next that the firm sets a nonlineartariff, and say its price for a consumer’s first unit is p1 while its price for a subsequent unitis p2. Then one can show that the profit-maximizing nonlinear tariff is p1 = 2, p2 = 1 ifλ < 1

2and p1 = 4, p2 = 2 if λ > 12 . Thus, when low-value users are more numerous, the

introduction of nonlinear pricing causes the low-value users to consume two units, whichis efficient. This results in a Pareto improvement compared to linear pricing. However,when high-value users are more numerous, the optimal strategy for the firm is to excludethe low-value users completely so as to fully extract the high-value surplus. Total welfaredecreases in this case compared with linear pricing, and all consumers are weakly worse off.In particular, this simple example shows that nonlinear pricing has an ambiguous effect onwelfare relative to linear pricing.While a consideration of when markets open or shut due to price discrimination does

offer some insights–in particular, the possibility of Pareto improvements is an uncontrover-sial benefit of price discrimination–this analysis does not take us very far. For instance,the welfare difference between a market being shut down and being open a tiny bit is notsignificant. Moreover, in practice markets are rarely completely shut down when price dis-crimination is banned. (In virtually all countries, there is someone who is willing to payalmost any price for a luxury car or a pharmaceutical product.) For this reason, in the nextsection we discuss the effect of price discrimination on total output.

5 The effect of price discrimination on output

A focus of earlier work on price discrimination has been on its effect on output. If differentproducts have different (marginal) prices when those products have the same marginal cost,then total output is sub-optimally distributed from a social welfare perspective. However, itmight be that output increases when firms are permitted to engage in price discrimination,and this effect might be sufficient to outweigh the undesirable “unequal marginal utilities”effect. Using this insight one may deduce that if price discrimination does not lead the firm(s)to expand output, it will cause welfare to fall.14 However, this does not tell us much in theevent that discrimination causes total output to rise, since welfare might then be higher or

14The effect of discrimination on total output was perhaps first studied by Robinson (1933). Schmalensee(1981) shows that if demands for the products are independent and marginal costs are constant, then if welfareincreases with discrimination it must be that total output increases. Varian (1985) extends this argumentto allow for cross-price effects, and Schwartz (1990) extends the argument to nonlinear cost functions.

10

lower. Thus, using changes in output as a test for the welfare effects of price discrimination iscapable only of delivering bad news. Moreover, outside some classroom examples, it is rarelyeasy to find conditions that characterize when output rises or falls with discrimination,and very detailed knowledge of consumer demand is needed to formulate accurate publicpolicy.15 However, if the firm is regulated the comparison is sometimes easier. For instance,Armstrong and Vickers (1991) and Armstrong, Cowan, and Vickers (1995) show that withcommon forms of average price regulation, total output necessarily increases if the firm ispermitted to engage in price discrimination. As already emphasized, however, this does notimply that price discrimination will be welfare improving.16

Another insight is that if the strong market is also the large market, then allowing pricediscrimination is often good for output and welfare. To see this, consider the followingexample.17 Consumers each wish to purchase a single unit of a monopolist’s product. Con-sumers either have a high valuation vH for the unit or a low valuation vL < vH . Thereare two markets, 1 and 2, and the fraction of the consumers in market i who have a highvaluation is λi. Suppose that λ1 ≥ λ2, so that market 1 has the greater proportion ofhigh-value consumers. Suppose that production is costless. If the firm is able to practicethird-degree price discrimination across the two markets, it will set a high price vH in marketi if λivH ≥ vL, and otherwise it will set the low price vL. If the firm cannot discriminate, itwill base its price on the overall fraction of high-value consumers across the two markets. Ifa fraction β of all consumers are in market 1, the overall fraction of high-value consumers isλ̄ = βλ1+(1−β)λ2. If the proportion of high-value consumers is similar in the two markets,so that the discriminating firm would like to set the same price in the two markets (eitherboth high or both low) then clearly policy toward discrimination has no impact on prices.However, if the distribution of consumers is quite asymmetric, in the sense that

λ1 >vLvH

> λ2 ,

the discriminating monopolist would set a high price in market 1 (the strong market) anda low price in market 2 (the weak market). In this case, policy which prohibits price dis-crimination will certainly have an effect on prices. If the weak market is the larger market,15When consumer demands are linear, and if no markets are opened as a result of price discrimination,

it is well-known that output is not affected by price discrimination, and hence that discrimination causeswelfare to fall. Schmalensee (1981, p. 245) concludes “If all demand functions are strictly concave or convexand the pi’s [the discriminatory prices] are not nearly equal, there is apparently no simple, general way totell if monopolistic discrimination will raise or lower total output.” However, see Cowan (2006) for moreprogress in this direction.16An interesting case involving price discrimination and total output is Competition Commission (1999).

In 1999, the supply of milk in the UK was fixed by European quotas. One distributor, Milk Marque, heldnearly a 50% market share, and was heavily engaged in price discrimination to its buyers. Since its totaloutput was fixed via the quota arrangements, this discriminatory pricing was likely to harm welfare. Thisfactor was an important part of the Competition Commission’s case against Milk Marque.17This is taken from Hal Varian, “A Big Factor in Prescription Drug Pricing: Location, Location, Loca-

tion”, The New York Times, 21 September 2000.

11

in the sense that λ̄ < vL/vH , the non-discriminating firm will choose to set a low uniformprice. This will increase consumer surplus and total welfare relative to the situation withprice discrimination. On the other hand, if the strong market is the larger market, in thesense that λ̄ > vL/vH , the firm will respond to a ban on discrimination by setting a highprice to everyone. This will harm consumers in the weak market, and lower total welfare.In sum, when a weak market–a market with relatively few high-valuation consumers–isalso a relatively small market, then price discrimination is likely to help consumers in thismarket and improve overall welfare. This suggests that a per se ban on price discriminationby country is too blunt a policy.Extending this monopoly analysis to situations of oligopoly does not usually make the

effect of price discrimination on output easier to predict, since one must consider firm-levelelasticities as well as market elasticities.18 One exception to this occurs in Armstrong andVickers (2001, section 4). Here the focus is on the competitive limit where prices are closeto marginal cost, and it is shown that if the weak market is also the market with lowermarket elasticity then total output falls with discrimination, and hence so does welfare.In competitive environments, firms might be forced to set a lower price in the “wrong”market (i.e., in the less elastic market) since firm-level elasticities might differ drasticallyfrom market-level elasticities. By contrast, with monopoly the firm generally sets high pricesin the correct (inelastic) markets.

6 When does price discrimination cause some prices torise and others to fall?

It is intuitive in many situations that when a firm is permitted to engage in price discrimina-tion some of its prices will fall while others will rise. That is to say, the non-discriminatoryprice is some kind of “average” of the discriminatory prices.Consider first the case of monopoly supply. Suppose a monopolist serves two markets,

1 and 2, which have independent consumer demands. The firm’s profit in market i when itsets the price pi in that market is denoted πi(pi). Then the profit-maximizing discriminatoryprices are characterized by π0i(pi) = 0, while the profit-maximizing uniform price p satisfiesπ01(p) + π

02(p) = 0. Except in the fluke case where there is no gain from discrimination, it

follows that in one market i we have π0i(p) < 0 and in the other market π0j(p) > 0. Assuming

profit functions are single-peaked it follows that if the monopoly can price discriminate itwill lower its price in market i and raise its price in market j.Matters are more complicated when there are competing suppliers. As emphasized in

Corts (1998), the chief difference with monopoly is that a market might be strong for onefirm but weak for its rival. In such cases price discrimination can cause all prices to fall, asdiscussed in section 7 below. However, when firms do not differ in their judgement of which

18See Holmes (1989), for example.

12

markets are strong, Corts shows that when price discrimination is permitted, prices will risein the market which both firms view as strong and prices will fall in the market which bothfirms view as being weak. (Corts uses the term “best-response symmetry” when firms agreeabout which market is strong and which is weak.) Armstrong and Vickers (2001, section4) and Armstrong (2006, section 3.2) provide simple examples of best-response symmetrywhere price discrimination causes some prices to rise and other to falls. In both examples,industry profit rises when discrimination is permitted, while consumer surplus falls.The discussion in this section has considered only the case of third-degree price discrim-

ination so far. Consider next a monopoly bundling example.19 There are two products, 1and 2, provided by a monopolist. Consumers are characterised by their valuations v1 andv2, where vi is a consumer’s valuation for product i. Suppose that her valuation for con-suming both products is just the sum v1 + v2. In general, the firm sets three prices: p1 isthe firm’s price for product 1 alone, p2 is the price for product 2 alone, and p12 is the pricefor the bundle of products 1 and 2. Figure 1 shows the resulting pattern of demand (whenp12 < p1 + p2).

......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

............................

0

v1

v2

p2

p12 − p2

p12 − p1

p1

Buy neither product

Buy onlyproduct 2

Buy onlyproduct 1

Buy both products

.........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

...................................................................................................................................................................

...................................................................................................................................................................

Figure 1: Pattern of Consumer Demand with Monopoly Bundling

Suppose the marginal cost of supplying either product is zero and consumer valuations(v1, v2) are uniformly distributed on the unit square [0, 1]2. If bundling is not possible (i.e.,

19Two key papers on monopoly bundling are Adams and Yellen (1976) and McAfee, McMillan, andWhinston (1989).

13

p12 = p1+p2), the profit-maximizing price for each product is 12 and the firm makes profit of12. If the firm is able to price discriminate by means of offering a discount for joint purchase(p12 < p1+p2), one can show the profit-maximizing prices are p12 = 43− 13

√2 ≈ 0.862 for the

bundle of two products and p1 = p2 = 23 for each stand-alone product. Thus the price for jointconsumption is lower than with unbundled pricing, while the price for a single item is higher.In this example at least, then, the market for joint consumption is the weak market, while themarket for individual consumption is the strong market. This pricing policy results in profitsapproximately equal to 0.55, which is 10% higher than the profit without bundling. Onecan calculate that aggregate consumer surplus with this mixed bundling policy is given by0.255, which compares to 0.25 with independent pricing. Thus, aggregate consumer surplusalso increases (slightly) when mixed bundling is used, although those consumers with strongtaste for one product but not the other are worse off with the bundling policy since theyconstitute the strong market. Again, then, we see that price discrimination is not necessarilyharmful to consumers or welfare, even in a monopoly context.

7 How price discrimination can intensify competition

Perhaps surprisingly, there are several situations where permitting firms to engage in pricediscrimination causes all prices to fall, so that competition is intensified. The examplesdiscussed here fall into two categories: models where firms discriminate on the basis ofconsumer brand preferences, and models of competitive bundling. We discuss these in turn.(Further situations in which price discrimination leads to tougher competition are found inthe dynamic context presented later in section 10.)Following the analysis of the previous section, consider cases where firms differ in their

view of which markets are strong and which are weak. Corts uses the term “best-responseasymmetry” for these cases.20 Take for example the model of Thisse and Vives (1988). Thereare two firms in a Hotelling market. Suppose first that firms can observe each consumer’slocation (or brand preference), and can price accordingly. A firm will then set a relativelyhigh price to nearby consumers in order to exploit those consumers’ unwillingness to travelso far to the other firm, so that its nearby consumers constitute a firm’s strong market. Inparticular, one firm’s strong market is the other’s weak market. In equilibrium, consumerspurchase the product from the preferred firm, which is efficient, and those consumers whoare almost indifferent between the two brands will obtain the best deal. If firms must set auniform price to all consumers, however, Thisse and Vives show that this uniform price canbe above all the discriminatory prices. Thus, when there is best-response asymmetry it ispossible that all prices decrease with price discrimination. In this model, price discrimination

20Nevo and Wolfram (2002) present evidence consistent with the hypothesis that price discriminationvia coupons in the breakfast cereal market exhibits best-response asymmetry, and that the introduction ofcoupons leads to a fall in all prices. They also document how firms allegedly colluded to stop the use ofcoupons.

14

has no impact on total welfare since all consumers just wish to buy a single unit, and theybuy this unit from the closer firm with either pricing regime.21 All consumers clearly benefitfrom price discrimination. Firms, however, make lower profits when they engage in this formof price discrimination compared to when they must offer a uniform price.22 As Thisse andVives (1988, page 134) put it:

“denying a firm the right to meet the price of a competitor on a discriminatorybasis provides the latter with some protection against price attacks. The effect isthen to weaken competition, contrary to the belief of the proponents of naive ap-plication of legislation prohibiting price discrimination like the Robinson-PatmanAct.”

The fact that firms might be worse off when they practice price discrimination is one ofthe key differences between monopoly and competition. Ignoring issues of commitment fornow (see section 10.1 below), a monopolist is always better off when it can price discriminate:the firm is free to choose a uniform price when discrimination is permitted but in general itis better off setting different prices. In the same way, an oligopolistic firm is always better offif it can price discriminate compared to when it cannot, for given prices offered by its rivals.However, as in many instances of strategic interaction, once account is taken of what rivalstoo will do, firms in equilibrium can be worse off when price discrimination is permitted.Firms then find themselves in a classic prisoner’s dilemma.We turn next to examples concerning bundling. The first such example is taken from

Matutes and Regibeau (1992), where two symmetric firms each offer a version of two prod-ucts. Consumers wish to purchase one unit of both products. (These two products could bepurchased from the same firm or from two different firms.) In a specific example, Matutesand Regibeau show that when firms are able to offer a discount for joint consumption, thefirms’ prices are uniformly below the equilibrium prices without bundling. Of course, eachfirm’s profit then falls when bundling is employed. Clearly, consumers are all better off as aresult of the price reductions caused by price discrimination. However, there is excessive loy-alty: because of the bundle discount too many consumers buy both products from the same

21As ever, one should be wary of reaching policy conclusions on the basis of these unit demand modelssince price levels have little role to play in welfare terms. If consumers had elastic multi-unit demands, pricereductions have a beneficial welfare impact. In addition, if firms have imperfect information about brandpreferences, price discrimination may induce some consumers to buy from their less preferred supplier. Forinstance, Bester and Petrakis (1996) consider a model where firms only know whether a consumer prefers itto its rival, but not by how much. In this model, firms set lower prices to those consumers who are knownto prefer the rival brand, with the result that those consumers who are almost indifferent between the twobrands will prefer to buy from the (slightly) less preferred supplier in return for the reduced price. A similarfeature appears in dynamic models of price discrimination in section 10.22Cooper, Froeb, O’Brien, and Tschantz (2005) use a variant of this model to argue that mergers are less

likely to be detrimental when firms practice spatial price discrimination than when they do not.

15

firm than is efficient, and so welfare falls with this form of discrimination.23 ,24 The economicreason why mixed bundling acts to intensify competition is rather subtle, and awaits furtherclarification. (It cannot be anything to do with best-response asymmetry, since firms do notview the various kinds of consumer as strong or weak in different ways.)The final examples in this section involve a multi-product firm facing a single-product

rival. I discuss two such examples. The first one involves pure bundling and the secondinvolves third-degree price discrimination. Consider first this bundling example.25 Supposethere are two products, 1 and 2, and each consumer potentially would like a unit of eachproduct. Firm A supplies both products and holds a monopoly over product 1, while firmB supplies a variant of product 2. Suppose consumers have additive utility for the twoproducts. Suppose consumers have homogeneous preferences for product 1, and all have thesame reservation value, v1, for this item. The two firms supply imperfectly substitutableversions of product 2, and if firm A sets the (unbundled) price pA2 while B sets the pricepB2 , firm A’s product 2 demand is q

A2 (p

A2 , p

B2 ) and B has demand q

B2 (p

A2 , p

B2 ). Firm A has

marginal cost cA1 < v1 for product 1 and marginal cost cA2 for its product 2, while B has

marginal cost cB2 for its product 2. Then A’s profit from product 2 is (pA2 −cA2 )qA2 (pA2 , pB2 ) and

B’s profit is (pB2 − cB2 )qB2 (pA2 , pB2 ). Given that A sets unbundled prices, its most profitableresponse to B’s price pB2 is denoted p

A2 = R

ASEP (p

B2 ) and similarly B’s best response to A’s

price pA2 is denoted pB2 = R

B2 (p

A2 ). When firm A sets unbundled prices for its two products,

the prices for product 2 are determined by the intersection of these two reaction functions,as depicted by γ on Figure 2. (Firm A sets the monopoly price pA1 = v1 for product 1.)Next, suppose firm A commits to sell its two products as a pure bundle, while prices are

determined in a second stage. Say that firm A’s price for the bundle is pA12. A consumerdeciding whether to buy from firm A or firm B knows that when she buys from A sheobtains an extra utility v1 due to the additional consumption of product 1. Therefore, firmA’s “effective price” for product 2 is pA2 = p

A12 − v1, and the demand for A’s bundle is just

qA2 (pA12 − v1, pB2 ). Firm A’s total profit is

(pA12 − cA1 − cA2 )qA2 (pA12 − v1, pB2 ) = (pA2 − [cA2 − {v1 − cA1 }])qA2 (pA2 , pB2 ) .23A variant of this model involves the two firms having to choose between separable pricing and pure

bundling. In this case, firms’ profit also falls in many cases when firms choose to sell their products only asa bundle, and typically if falls by more than when mixed bundling is employed. See Matutes and Regibeau(1988) and Economides (1989) for this analysis.24The example analyzed in Matutes and Regibeau (1992) is extended in Armstrong and Vickers (2006)

to allow for asymmetric products, non-uniform distributions, correlation in brand preferences, and shoppingcosts for purchasing from more than one supplier. They show in this more general framework that mixedbundling continues to harm profit and welfare, and to boost consumer surplus, relative to linear pricing.However, they also extend the bundling model to allow for elastic, multi-unit demand for each product,in which case the impact of price discrimination on profit, welfare and consumer surplus is shown to beambiguous.25This is essentially Example 2 in Whinston (1990). For more detailed discussion of the Whinston model

and the literature that follows, see Neven (2007).

16

This is similar to its profit from product 2 when there is unbundled pricing, except thatits cost cA2 is shifted down by {v1 − cA1 }. Therefore, firm A’s most profitable effective pricepA2 , given B’s price p

B2 , which is denoted p

A2 = p

A12 − v1 = RABUND(pB2 ), is shifted downwards

compared to the case of unbundled pricing. The (effective) prices for product 2 are nowlocated at β on Figure 2, which are lower for both firms. Typically, both firms’ profits fallwhen firm A offers its products as a bundle. Therefore, when there is an existing firm inmarket 2–so that entry deterrence is not an issue–in this particular model we would notexpect the multi-product firm to choose to sell its products as a bundle.

................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

............................

pA2

pB2

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

...

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

............................................

...

...................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................

..........................................................................................................................................................

.......

.......

.......

.......

.......

.......

.....

RASEP (pB2 )

RABUND(pB2 )

RB2 (pA2 )

β

γ

Figure 2: The Strategic Effect of Pure Bundling

Why does bundling by the multi-product firm lead to more intense competition? Asexplained by Whinston (1990, page 884), when firm A bundles its products, in order tomake profitable sales of its monopolized product 1, it must also make sales of its product2. This leads it to cut price in an effort to take sales away from firm B, an effect Whinstonterms “strategic foreclosure”. However, whether this strategic foreclosure effect operates ornot depends on the details of the model, as we will see in section 8.A closely related model, involving third-degree discrimination instead of bundling, is

the following.26 Suppose firm A serves two independent markets, 1 and 2. Market 1 is26See Armstrong and Vickers (1993) for formal analysis. This paper also argues that the effect of allowing

17

monopolized by firm A, whereas in market 2 there is a rival firm B. If firm A can pricediscriminate in the sense that it may offer selective price cuts and set distinct prices in thetwo markets, it will set the monopoly price in market 1, while prices in market 2 will bedetermined by the competitive interaction between the two firms. On the other hand, if firmA cannot set different prices in the two markets it will have to trade off the benefits of settinga low price in the competitive market versus a high price in the captive market. In particular,if firm A prefers to set a higher price in the captive market than in the competitive market,so that the former is firm A’s strong market, the effect of a ban on price discrimination is toreduce the price in the captive market and to raise prices in the competitive market. Theprofit of the single-product rival clearly increases when such a ban is imposed, while theeffect on firm A’s overall profit is not clear-cut.27

8 How discrimination can relax competition

There are a variety of ways in which price discrimination can act to relax competitive pres-sures in the markets, to the detriment of consumers.28

When a multi-product firm faces a single-product rival, it is possible that bundling can actto reduce the intensity of competition.29 To see this, suppose again there are two products,1 and 2. Firm A supplies both products while firm B can supply only product 2. The twofirms here offer identical versions of product 2. A consumer has valuation v1 for product 1and v2 for product 2 (and valuation v1 + v2 for both products). Suppose the firms’ costs ofsupply are normalized to zero. For simplicity, suppose that (v1, v2) is uniformly distributedon the unit square [0, 1]2. Then, if firm A sets unbundled prices for its two products it willset the monopoly price pA1 =

12in its captive market. All profits are competed away for

product 2, and that product’s price is equal to zero. Therefore, with unbundled pricing firmA’s profit is 1

4, while firm B makes nothing.

Suppose instead that firm A commits to bundle its two products together, prior to thetwo firms choosing prices. Firm A sets a price pA12 for its bundle, while firm B sets the pricepB2 for its product. Figure 1 depicts the pattern of consumer demand (where p

A1 =∞). By

calculating the areas in the figure, one can calculate the equilibrium prices are approximately

price discrimination on the intensity of competition is exacerbated when the multi-product firm is regulatedand operates under an average-price constraint. (If it reduces its price in the competitive market it can thenraise its price in the captive market.)27Dobson and Waterson (2005) present a related model where a national retailer operates in a number

of markets, in some of which it is the sole supplier and in the remainder of which it faces a single localcompetitor. They show that it is possible for the chain store to benefit if it commits to a national pricingpolicy (i.e., if it does not price differently depending on competitive conditions in each local market).28See Buccirossi (2007) for a fuller account of “collusive” practices.29This bundling example is taken from Section III.E of Nalebuff (2004). See section II of Whinston (1990),

Carbajo, De Meza, and Seidman (1990) and Chen (1997a) for earlier analyses.

18

pA12 ≈ 0.61 ; pB2 ≈ 0.24. Therefore, all prices rise compared to the situation in which firmA sets unbundled prices. In effect, bundling helps to differentiate the two firms’ offerings:firm A offers a superior “product” (by virtue of its additional product 1 in the bundle). Therespective profits of the two firms are approximately πA ≈ 0.367 ; πB ≈ 0.067, which arealso both higher compared to when firm A prices its products separately. This example, incombination with the discussion of theWhinston model in the previous section, indicates thatthe effect of bundling on a multi-product firm’s incentive to be aggressive to single-productrivals is ambiguous and depends on the fine details of the model of consumer demand.Without very detailed data on consumer preferences, it is hard for an anti-trust authorityto predict a priori whether bundling will relax or intensify competition.Another way in which price discriminationmight relax competition is when price-matching

contracts are used, i.e., when a firm promises consumers it will match a lower price of a ri-val firm if consumers can find evidence of such a price. Such contracts are a form of pricediscrimination since the price a consumer pays differs according to their knowledge of otherfirms’ prices and their willingness to go to the trouble to provide evidence of a lower price.If consumers are well informed about all rivals’ prices and there are no effort costs involvedin documenting the lower prices, there is a clear danger of collusion: there is no incentivefor one firm to undercut another since the low-priced firm will not obtain greater marketshare and will simply lower its and its rivals’ prices. In this stark framework, public policywhich prohibits the use of such “competitor-based” discrimination schemes would help re-store normal competitive pressure in the market, to the benefit of consumers. However, ifconsumers face costs of searching for price offers or providing evidence of low prices, a moreconventional price discrimination motive emerges: sorting consumers on the basis of theirvalue of time (akin to the use of discount coupons). In such cases, the collusive impact ofcompetitor-based discrimination is blunted, and consumers might sometimes be better offwith this form of price discrimination is employed.30

A further way in which price discrimination might act to limit competition occurs whenit leads to negative network effects. Consider an telecommunications market in which sub-scribers join one network or another. Public policy might or might not permit firms to pricediscriminate on the basis of the destination network. This form of discrimination is presentif a subscriber on network A faces a different call charge if his call is made to another sub-scriber on network A or to a subscriber on a rival network B. If firms each make it cheaperto make calls to people on rival networks than to people on their own network, then negativenetwork effects are present and, all else equal, a subscriber prefers to join a smaller networksince they can then make a greater fraction of their calls at the cheap rate. It is well knownthat markets with negative network effects of this form will not be very competitive, since afirm which offers a low price will not attract many new subscribers. Therefore, this form ofprice discrimination can act to relax competition, to the detriment of consumers.31

30See Corts (1996), for example.31I have skated over the means with which firms agree to set lower prices to make calls to subscribers on

19

9 Effects of discrimination on entry

The effects of price discrimination on entry is a central concern of competition policy. Threebroad issues are discussed: (i) the effect on the equilibrium number of firms in a monopo-listic competition framework; (ii) the effect on a monopolist’s incentive to enter an adjacentmarket; and (iii) the effect on a potential entrant’s incentive to enter a monopolized market.

9.1 Free entry

Price discrimination can raise or lower equilibrium profits in a market with a given numberof firms, as seen in the previous sections. When there is free entry of firms, it follows thatthe equilibrium number of firms will rise with price discrimination in the former case, butwill fall in the latter case when more intense competition will drive out some firms. In somecommon models of oligopolistic interaction–though by no means all–it is well known thatfree entry will result in too many firms entering from a total welfare perspective.32 In thosecases where price discrimination raises profits, the resulting greater entry will only exacerbatethe welfare costs of excessive entry. By contrast, when price discrimination destroys profits,the excess entry problem will be mitigated (although typically not overturned).33

9.2 A monopolist’s incentive to enter an adjacent market

The second topic in this section involves a monopolist in one market entering an “adjacent”market. The motives considered here are not necessarily anti-competitive–there need notbe a motive to drive firms out of the adjacent market–but rather the firm wishes to engagein price discrimination, and in some situations the only way it can do this is by participatingin the adjacent market.

Ability to offer quantity discounts: In this example, product 1 is monopolized by firm A,while product 2 is potentially supplied in a competitive market with price equal to cost. Forsimplicity, suppose all production is costless. Suppose product 2 is a complementary product

rival networks. In the telecommunications context, this could be done via low “call termination” rates. Iffirms agree to set their charge for delivering calls from rival networks below the associated cost, then firmswill have an incentive to set low charges for calls made to rival networks. See Laffont, Rey, and Tirole (1998),Gans and King (2001) and Berger (2005) for further details.32See Mankiw and Whinston (1986), for instance.33See Borenstein (1985), Norman and Thisse (1996) and Bhaskar and To (2004) for further analysis of the

effect of price discrimination on the free-entry number of firms. Ordover and Panzar (1982) investigate theeffects of an upstream monopolist’s use of two-part pricing on the number of downstream firms. They findthat in general the monopolist will not set a socially optimal two-part tariff, despite the fact that it knowseverything about its downstream customers. Thus, in this respect there is a contrast between selling to finalconsumers (when the use of two-part tariffs leads to an efficient outcome, as discussed in section 2 above)and selling to intermediate customers.

20

to product 1, but consumed in variable quantities by consumers. Consumers wish to buyjust one unit of product 1, however. (The classical examples of this situation are computersand punch-cards, or photocopiers and toner.) Consumers differ in how much of the jointproduct they wish to consume. Suppose demand for the joint product by a consumer withdemand parameter θ is θ−p2 if the marginal price for product 2 is p2. Suppose θ is uniformlydistributed between 0 and 1. If p1 is the price for product 1, a consumer’s net surplus is12(θ − p2)2 − p1. Therefore, a consumer will participate in the market if θ ≥

√2p1 + p2, and

total demand for product 1 and product 2 is respectively

q1 = 1−³p

2p1 + p2´; q2 =

12(1− p2)2 − p1 .

If the monopolist for product 1 ties the purchase of its monopoly product to purchaseof the potentially competitive product 2, it can set a price above cost for product 2. Itwill then choose the pair of prices (p1, p2) to maximize total profit p1q1 + p2q2, and thisentails p1 = 0.08 and p2 = 0.2. This brings in profit of 0.08, while consumer surplus isapproximately 0.043. In particular, the monopolist wants to set a price above cost forproduct 2.34 If the monopolist did not tie product 2 with product 1, all active consumerswill buy product 2 on the competitive market with price p2 = 0. This will surely bringthe monopolist less profit than the tying strategy (since it could choose p2 = 0 under thetying regime, but preferred to set p2 = 0.2). Therefore, the monopolist has an incentiveto “leverage” its monopoly position for product 1 into the market for product 2. If suchtying is not permitted (so p2 = 0), the monopolist would significantly increase its product 1price to p1 = 0.22. Its profit is then 0.074, while consumer surplus is essentially unchangedat 0.044. In this example, then, total welfare is higher if the monopolist is permitted totie its product, although in general the comparison is ambiguous.35 ,36 Notice also that themonopolist would like to pursue this tying strategy even it incurred a slightly higher costin supplying product 2 than did the competitive rivals. (This assumes that the monopolist

34The monopolist will do even better if its sets a fully nonlinear tariff, rather than just a two-part tariff,for consumption of the joint product.35See section 3.3.1.5 in Tirole (1988) for a clear account.36In fact, the firm can also do well if it offers consumers an optional bundled tariff. Suppose that the

firm offers consumers the profit-maximizing unbundled tariff (p1 = 0.22, p2 = 0). Suppose, in addition tothis tariff, the firm offers consumers the option of another tariff which has a lower price for the monopolycomponent but a higher price for the competitive good. Such a tariff will be attractive to consumers withrelatively low demands. Specifically, if the firm offers the optional tariff with p̂1 = 0.037 and p̂2 = 0.272 thenthose consumers with 0.54 ≤ θ ≤ 0.82 will use this new tariff. (Some of these are new consumers and someare “cannibalized” from the existing tariff.) Since consumers have the option of using the unbundled tariff,consumer surplus surely rises with the addition of this tariff. Also, the firm’s profit with the new optionaltariff is 0.081, which is higher than that generated by the unbundled tariff on its own, and also higher thanthat generated by the optimal single two-part tariff (p1 = 0.08, p2 = 0.2). Therefore, the introduction ofthis optional tariff represents a Pareto improvement compared to when the unbundled tariff is used on itsown. See Nalebuff (2005) for related analysis (where the same model for consumer demand is employed).

21

cannot monitor a consumer’s consumption of product 2 from rival suppliers.) In this sense,the monopolist has an incentive to exclude more efficient rivals.

Ability to offer bundling discounts: Since a multi-product monopolist typically gains whenit can bundle its products–see the discussion in section 6 above–it follows that a multi-product monopolist makes greater total profit than if the same products were supplied by aseries of single-product monopolies. (If the products were supplied by single-product firms,then prices would be product-by-product.) This is true even when the various products areneither complements or substitutes in the usual sense. In technical terms, there may be“informational economies of scope” involved in multi-product supply, even if more familiareconomies of scope are not present.37

To illustrate, return to the example in section 6 where there are two products, 1 and 2, andconsumers have valuations (v1, v2) for these two products individually and valuation v1 + v2for joint consumption. If production is costless and the valuations are uniformly distributedon [0, 1]2, the multi-product monopolist obtains profit of 0.55 by practising mixed bundling,while a pair of single-product monopolies each supplying one product make a combined profitof 0.5. To see how this affects a monopolist’s incentive to enter an adjacent market, considerthe following broadcasting scenario. There is an incumbent broadcaster who currently holdsbroadcasting rights to a particular sports event. The rights to a second, perhaps unrelated,sports event then become available for auction. Because the incumbent broadcaster will beable to engage in bundling if it obtains the second rights, but a new broadcaster will not,the incumbent will be willing to pay more for the new rights than a new broadcaster. (Itwill be prepared to pay up to 0.3 for the new rights, whereas a new broadcaster could onlypay 0.25.)Thus, even if for no other reason, this argument suggests that an existing rights holder is

likely to win a bidding war for any new rights. Indeed, in a richer model where the entrantmight offer a superior service (e.g., it has lower supply costs or a higher quality service),the incumbent might be prepared to pay more for the additional rights than the superiorentrant. However, set against this is the fact that consumers often benefit from bundling.For instance, in this particular example, consumers are better off if the incumbent wins thenew rights and practices bundling. Thus, if policy-makers are deciding whether to permitbundling, they may have to trade-off the possible consumer benefits of bundling againstthe possible foreclosure of more efficient entrants. The information required to implementaccurate policy in this environment is formidable.This provides a simple example of “conglomeration” effects (or “portfolio” or “range”

effects), which have risen in prominence in anti-trust cases in recent years. See Neven(2007) for a more detailed account of such cases, which include the controversial General

37See Bakos and Brynjolfsson (2000) for further analysis of this issue. Of course, the argument does notapply to third-degree price discrimination, where there are several separate markets and each consumerwishes to purchase just a single product from the multi-market firm.

22

Electric/Honeywell merger. One reason why these cases are challenging for the competitionauthorities is that the standard approaches to market definition no longer always applywell.38

9.3 An entrant’s incentive to enter a monopolized market

One of the most controversial aspects of price discrimination is the possible impact of anincumbent’s ability to price discriminate on the incentives for entry. There are a number ofways in which price discrimination by an incumbent firm affects the incentives to enter itsmarkets. Constraining the incumbent’s freedom to choose its prices affects its response toentry, and hence the expected post-entry profits of the entrant.This is illustrated in the context of the model of third-degree price discrimination based

on Armstrong and Vickers (1993) discussed in section 7. There is a multi-market incumbentfacing potential entry into one of its markets. If the incumbent is permitted to set differentprices in its two markets, then it is plausible that it will react more aggressively to entrycompared to when it must set the same price in the competitive and the captive market. Weconclude that the post-entry profit of the entrant is likely to be lower when the incumbentcan price discriminate than when it cannot. If the entrant has a fixed cost of entry, it willenter only if it expects post-entry profit to cover its entry cost. There are then three cases toconsider. If the entry cost is large, there will be no entry regardless of whether the incumbentcan price discriminate. In this case, the social desirability of price discrimination is exactlyas in the standard monopoly case, and this is ambiguous in general. Similarly, if the entrycost is very small, entry will take place regardless of policy towards price discrimination.The interesting case is when the cost of entry lies in the intermediate range where entry isprofitable only if the incumbent cannot price discriminate, so that a ban on price discrimi-nation acts to induce entry. In such cases it is plausible that a ban on price discriminationwill cause the prices in both markets to fall: if discrimination is possible, there will be noentry and the incumbent will charge monopoly prices in each market; if the incumbent mustcharge a common price in the two markets, this will bring in the entrant and force both ofthe incumbent’s prices down from monopoly levels.The general principle, as in the Thisse-Vives quote in section 7 above, is that denying

an incumbent the right to meet the price of a competitor on a discriminatory basis providesthe latter with some protection against price attacks. While the effect of a ban on pricediscrimination is indeed to weaken competition if the entrant is already in the market, oncethe ex ante incentives to enter are considered, the effect of a ban on price discrimination mightactually be pro-competitive. However, the welfare effects of a ban on price discrimination

38Another way in which price discrimination interacts with market definition is that, (i) when a hypothet-ical monopolist is able to price discriminate between two or more consumer groups, these consumer groupswill make up separate anti-trust markets, while (ii) when the monopolist cannot price discriminate, thewhole market becomes the relevant market for anti-trust purposes. For instance, see the Horizontal MergerGuidelines issued by the United States Department of Justice and the Federal Trade Commission, 1992.

23

in this context are not clear cut. For instance, since the incumbent is reluctant to cut itsprofits in the captive market by meeting its rival’s price in the competitive market, even ahighly inefficient entrant might prosper. While preventing an incumbent from engaging inselective price cuts is likely to be a powerful means with which to assist entry, as with manyforms of indirect entry assistance the danger of inefficient entry is never far away.This question of whether to permit (above cost) selective price cuts is of substantial

importance in practice. For instance, many network industry incumbents (such as privatizedtelecom companies) are required to offer geographically uniform retail tariffs, even thoughthey face competition only in particular areas (such as cities). While such a policy will indeedact to encourage entry and to “bring the benefits of competition to the whole country” (as itis often put), it is also likely to induce inefficient, or excessive, entry into the attractive marketsegments. The question is also important in markets without a history of regulation. Forinstance, Vickers (2005, page F249) summarizes a recent anti-trust case involving shipping:

“Above-cost price cuts were at issue in the case of Compagnie Maritime Belge,on which the ECJ gave judgment in 2000. The enterprise, which had a near-monopoly position on certain shipping routes between Europe and West Africa,had selectively cut prices to match those of its competitor, though not demon-strably to below total average cost. The Court saw the risk that condemningsuch pricing could give inefficient rivals a safe haven from the full rigours ofcompetition, but in the circumstances at hand judged that there was abuse (al-beit not abuse under the heading of predation) because the selective price cutswere aimed at eliminating competition while allowing continuing higher pricesfor uncontested services.”

The situation in which a firm selectively lowers its prices in more competitive markets isclosely connected to the situation in which a firm selectively offers low prices to customers ofa rival. We will discuss this issue in more detail in section 10, albeit mostly in the context ofcompetition between symmetric firms. When firms are symmetric, we will see that selectiveprice cuts to a rival’s past customers can reduce prices to all consumers.A second way in which price discriminationmight impede entry concerns (pure) bundling.39

In section 7 we presented Whinston (1990)’s model where, if a multi-product firm bundlesits two products, the equilibrium profits of the single-product rival are reduced. Therefore,if the multi-product incumbent can commit to bundle its products together before the entrydecision has been made, this can act to deter entry. (Again, this will depend on the sizeof the fixed cost of entry.) Once entry is deterred, the incumbent can raise its bundle priceto the monopoly level. There are a broad range of situations in which the incumbent finds

39The most prominent case relevant to this is the Microsoft case, and in particular the part of the caserelating to the bundling of the Internet Explorer browser with its Windows operating system–see section7.5 of Motta (2004) for some further details. An earlier relevant case from 1979 is Berkey Photo v. EastmanKodak.

24

it privately profitable to deter entry in this way, but where total welfare would be higherif entry took place. Such instances provide a coherent case for a ban on bundling by anincumbent.From the incumbent’s point of view, one problem with the strategy is that, if entry does

take place, the incumbent also makes lower profits with bundling than with unbundled prices.That is to say, its plan to bundle products might not be credible. The entrant might predictthat the incumbent will renege on its promise to bundle if entry does take place, in whichcase the threat to bundle carries little weight. However, there are plenty of cases where thebundling decision is a long-term decision (if it is built into the product design, for instance),in which case it has credibility and the ability to deter entry.40

Another problem is that apparently minor changes to the model imply that bundling,even if credible, will not deter entry. For instance, in section 8 we saw that if consumers hadheterogeneous valuations for the monopolized product, it might be that both firms’ profitsare higher if the multi-product firm bundles its products. In this case, bundling acts toencourage entry. Since the incumbent would still like to deter entry so it can set monopolyprices in both of its markets, if there is a fixed cost of entry the incumbent might wellprefer to commit to unbundled pricing to deter entry (even though if entry did occur, theincumbent would prefer to revert to the bundling strategy which boosts both firms’ profits).While a monopolist would prefer to bundle its products in this example, unbundled pricingwith monopoly yields higher profit (0.5) than does bundling with entry in one market (whichyields the incumbent profit of only 0.367).This discussion of bundling so far has assumed the two products enter additively into

consumer utility: the fact that a consumer has purchased product 1 has no impact on herwillingness-to-pay for product 2. Many examples of bundling involve complements. In theextreme case of perfect complements, a consumer places no value on product 2 unless shealso has product 1. The analysis changes somewhat when the two products are perfectcomplements. For one thing, a firm with a monopoly on one component clearly has theability to foreclose entry into the complementary product market, simply by bundling itstwo components together. Whether it has the incentive to do this is another question.Whinston (1990, section III) shows in a benchmark model that the incumbent has no suchincentive. Indeed, the incumbent generally benefits from the presence of a differentiatedentrant in one component market. For instance, if the entrant provides an identical product2 component but at lower cost, the incumbent will be able to extract all the entrant’s profitby setting a product 2 price below its own cost, and raising its product 1 price accordingly.This is an instance of the Chicago School’s insight that an incumbent has no incentive toprevent entry by a more efficient firm in a complementary product market.41 Similarly, if the

40Nalebuff (2004) mainly considers a model in which the incumbent firm not only makes its bundlingdecision prior to entry, but decides on its prices before entry. Among other results, he shows that bundling isan effective way to deter single-product entry if the incumbent does not know in advance into which marketthe entrant will enter.41The Chicago argument relies on the integrated firm possessing accurate information about its rivals’

25

entrant offers a differentiated product 2 component which appeals more to some consumersthan the incumbent’s own product, the incumbent will make more profit by allowing theentrant to deliver its monopoly component to the entrant’s own pool of consumers.42

Still in the context of perfect complements, Choi and Stefanadis (2001) examine thecase where an entrant must innovate if it is to produce a successful substitute for one ofthe incumbent’s components. If the incumbent bundles its two product, this implies thatthere is no scope for single component entry. Entry then requires successful innovation onboth fronts. Bundling can therefore reduce the incentive to innovate and so can increase theprobability that the incumbent retains its markets. However, it is not always optimal forthe incumbent to bundle, since it benefits when an entrant successfully innovates on just onecomponent (as in the previous paragraph). Carlton and Waldman (2002) present a modelwhere entry into the monopolized market is more likely to occur if the entrant is first active inthe competitive market. If the entrant offers a superior product in the competitive market, orif its costs are lower there, then entry there is socially desirable. If entry were to be limitedto that market, the incu